Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!crdgw1!uunet!mcsun!ukc!tcdcs!dce.ie!em
From: em@dce.ie (Eamonn McManus)
Newsgroups: comp.text.tex
Subject: Redesigning TeX (was: Why use TeX if ...)
Message-ID: <unhcopy@dce.ie>
Date: 14 May 91 21:00:33 GMT
References: <1991May9.204113.17636@beaver.cs.washington.edu> <1991May10.065219.23433@agate.berkeley.edu> <1991May10.211802.4344@csrd.uiuc.edu>
Organization: Datacode Communications Ltd, Dublin, Ireland
Lines: 46

In my opinion the TeX language, as opposed to the implementation, is not
as neatly designed as it might be.  (I'll leave comments on the
implementation for another day.)  It's not true to say that TeX has no
syntax to speak of.  Primitives should have been designed where possible
to have a syntax that could be described by macros, if only to the
extent of surrounding their arguments with braces.  It's ridiculous that
there should be constructs like
    \hskip 1pt plus 3fil
where TeX determines pretty much by magic where the argument ends.  (And
if the next word happens to be `minus' the unwary user is in for a nasty
surprise.)  I see no reason why this could not have been
    \hskip{1pt plus 3fil}
or, eliminating dimension keywords completely and using the implicit
multiplication feature,
    \newdimen\pt \pt=65536\sp
    \hskip{1\pt plus 3\fil}
There are numerous other syntactical complexities: consider \halign,
\hrule, \hbox, \leaders, \let, etc.  I'm not saying that all of these
should be shoe-horned into a brace-delimited form, but there is
certainly more complexity in the syntax than necessary.

In discussing the number of primitives, I don't think it is meaningful
just to count the number of asterisks in the TeXbook's index.  Some
primitives are just parameters, like \parindent and \maxdeadcycles and
\abovedisplayshortskip.  Others are easily expressible in terms of other
primitives, and are either there for efficiency (this is presumably the
case for \hfil etc) or because they never got removed (there doesn't
seem to be a good reason for \gdef given \global).  It would be nice to
tidy up the primitives a bit: do we need both \raise and \lower?
Shouldn't we have a parameter \interaction instead of \nonstopmode etc?
If we didn't have \futurelet could we make do with \let and
\afterassignment?  Would it be a terrible thing if \gdef and \above and
\hfilneg were macros in plain.tex instead of primitives?

One other thing I would like to change is the builtin limit in the
language design of having 256 of each sort of register.  It would make
much more sense (though perhaps at the expense of efficiency) if
\newdimen etc were the primitives and we forgot about \dimen.

Having bitched so much I must say that I think that TeX as it stands is
a wonderful system and its design, despite my complaints, is excellently
done.  Having something as portable as the existing system is much more
important than ironing out a few design nits.

,
Eamonn